Kitz Forum
Internet => General Internet => Topic started by: aesmith on September 21, 2021, 09:04:32 AM
-
Hi,
Just posting this after some messing around on Saturday. At home we use LTE as the main Internet connection, with ADSL as backup (plus a few things preferentially routed over ADSL). ADSL is around 3.5meg download and 800k upload. On Saturday there was a local LTE outage, taking out Three from both masts within range and incidentally also taking out EE but not O2.
What struck me was how many things simply did not work on the ADSL. For example some BBC news pages would time out. Three user account login returned an error. Google Drive synch threw the most enormous sulk, even the preferences menu was greyed out. Deleting all settings it simply wouldn't connect to the Google account. During this time the ADSL wasn't maxed out, although it was carrying traffic. So I am a bit puzzled as to what was actually happening, what were these applications and web pages doing that causes an almost immediate error on a 3.5meg connection?
If this is really the case then it's clearly a change over the last few years as we used to run everything over that same ADSL, with perfectly acceptable performance given the speed.
I would be interested in comments or thoughts. It would be disappointing if people are now creating web sites and services that simply won't work without at least a five or 10 meg connection. Just locally there are a lot of places in Aberdeenshire where you are lucky to even get 2meg.
-
No, they definitely haven't, I can't even think how you'd do that as a website doesn't know how fast your broadband is.
What DNS server are you using in the router? Setting it manually to Cloudflare or Google if not already might help, as it could be a glitch with your ISPs DNS server causing this whereas using the Three DNS server was fine.
-
Cheers. I was wondering about DNS but I can't see anything that would mess it up. DNS is served by the router, so all DHCP devices have the router IP as their DNS server. In turn the router uses the Opendns (now Umbrella) servers 208.67.222.222 and 208.67.220.220. As far as I can determine the cached entries held by the router are just host to IP mappings, not routes. Certainly the few traceroutes that I ran all went the correct way, to the DSL router and out to AA, rather than to the LTE router and failing.
I think I need to run some tests if I can agree a downtime window. Disabling the LTE interface in the router will simulate the fault most accurately.
Thinking about Google Drive Sync - I wonder if it inherently uses Google DNS servers instead of those that the host is configured to use? I don't at the moment have the router set to over ride that, so any DNS preferences used by any host will bypass the router's cache.
-
Designers can definitely cause issues when using a slow connection.
There's a difference between "doesn't work" and "unusable".
The problem is that website creators have so much crap such as tracking included in the pages that on a slow connection they are unusable (and may time out).
-
That's a nice distinction and on that basis I'd say BBC news was "unusable" whereas Google Backup & Sync "didn't work". Actually that Google thing makes me a bit cross, why on earth would they disable the menus even if it was completely disconnected from the Internet? It just makes all their recommended troubleshooting steps meaningless.
-
The problem is that website creators have so much crap such as tracking included in the pages that on a slow connection they are unusable (and may time out).
Honestly, I get that fairly regularly on a GOOD connection. Googles advert servers has days where they barely work sometimes. Not super often, but often enough to be annoying.
That's a nice distinction and on that basis I'd say BBC news was "unusable" whereas Google Backup & Sync "didn't work". Actually that Google thing makes me a bit cross, why on earth would they disable the menus even if it was completely disconnected from the Internet? It just makes all their recommended troubleshooting steps meaningless.
The menu disabling is probably down to AJAX or the Javascript itself not loading properly. After all, modern websites have a lot of functionality that runs in the browser itself so the whole page can load while the interactive functionality may be hitting a completely different server.
-
The menu disabling is probably down to AJAX or the Javascript itself not loading properly. After all, modern websites have a lot of functionality that runs in the browser itself so the whole page can load while the interactive functionality may be hitting a completely different server.
Google Backup and Synch is a local application, not a web page. I'm away to test on Windows to see firstly whether it talks specifically to Google DNS, and how it behaves if the network is completely down.
-
Not quite the same test, but starting Windows with network disconnected, the Google B&S menu has preferences greyed out as well.
-
Can I have an:
export hide-sensitive
From your rb4011 please.
-
That does sound strange , up here in the sunny north I don't have any trouble loading pages ( although they are not as quick as they could be ) on a 3meg ADSL . Plus we can stream stuff just not in HD
-
Can I have an:
From your rb4011 please.
Sure, but I'll send it by PM if you don't mind. The "hide sensitive" only obscures passwords, all other information is retained.
-
Sure, but I'll send it by PM if you don't mind. The "hide sensitive" only obscures passwords, all other information is retained.
Yes, I've noticed that. I see on the latest routeros you can run containers, not tried yet.
-
You'll need to bear with me, the exported text exceeds the max PM size. Would any subset of the config be any use?
-
Before I had multiple lines, I had just under 2 Mbps download on ADSL, and everything worked fine; in fact everything worked fine using 49k dialup. I think you need to get AA involved in diagnosing this. Maybe you’re getting packet loss?
-
You'll need to bear with me, the exported text exceeds the max PM size. Would any subset of the config be any use?
Firewall and routing, maybe DNS
-
Firewall and routing, maybe DNS
PM sent, firewall, DNS and ip routes. Happy to answer any questions. The guts of the failover is in the route table, the recursive routes and associated gateways.
-
Before I had multiple lines, I had just under 2 Mbps download on ADSL, and everything worked fine; in fact everything worked fine using 49k dialup. I think you need to get AA involved in diagnosing this. Maybe you’re getting packet loss?
Two and a half years ago we ran everything on this same ADSL without problems. In actual fact it runs faster now than it did back then as AA has nailed it up to disable DLM. Previously it used to keep getting slowed down, and on 20CN just one reconnection a slow speed took four days before the profile reset. So it's different behaviour now.
I think I need to properly test myself before I have anything that AA could comment on. Shutdown the LTE and check all the routing converges correctly with nothing pointing the wrong way. As another test I have an SSID directly from the ADSL router so I can test completely independently of anything else.
-
One thing that springs to mind, is it sites you were already using that stop working? I know on pfSense there is an option for sticky states (I can't remember the exact name) which sets up the NAT session so once you hit a site, it will always use the same gateway. You'd expect the router to wipe the NAT states for a gateway that is down, but maybe it isn't doing?
This may be completely barking up the wrong tree as that option is usually for load balancing rather than failover, but it seemed worth considering if there might be a similar option on your router.
-
The Mikrotik routers do something similar which is a functional defect. If my router was doing NAT and the route changed, it would retain the old NAT until it closed or timed out. That's not a problem with TCP, or infrequent UDP where it times out anyway. But it was a problem with SIP where the UDP by design has to be frequent enough to keep NAT and firewall entries in place. I did not resolve that, although I had some sort of plan for a script that would clear sessions when a gateway changed state.
In the current design the 4011 doesn't do NAT for either LTE or ADSL, so that specific issue doesn't apply any more. There may be some other hang up that's I've not yet found.
One thing to explore is redirect. I don't at the moment have a L3 transit to the ADSL router, so main router is default gateway and .1 on the subnet, ADSL is .2 on the same subnet. This may redirect, or it may route I have not yet checked or tested.
-
That does remind me, last time I yanked out DSL the SIP did fail to transition over. I do need to figure out how to fix that at some point.